CRAN firedrill 2023-05

[jcheng5]

• clang-ASAN, gcc-ASAN: Fix use after free (Fix use-after-free error #363)
• clang-ASAN: "runtime error: applying zero offset to null pointer" #365 (Fixed in Fix 'applying zero offset to null pointer' issue #366)
• Mac: " Assertion failed: (handle->flags & UV_HANDLE_CLOSING), function uv__finish_close, file core.c, line 258." (Fixed by the other memory fixes)
• Compile libuv with -DNDEBUG to prevent assertions
• Declare autoconf, glibtoolize, etc. in SystemRequirements (I think we should explain that these are optional)
• Use system copy of libuv if found #368
• Don't dynamically link to libuv on macOS #374
• Put libuv copyright holders in Authors@R

---

1.6.11 release checklist here: #371

[jcheng5]

I looked into the -DNDEBUG thing, messing with the environment variables that are set for the ./configure call in src/Makevars.

1. CPPFLAGS="$(CPPFLAGS)" doesn't fix it, as -DNDEBUG is not part of it.
2. CPPFLAGS="$(ALL_CPPFLAGS)" sort of fixes it--it puts -DNDEBUG in the right place, but using the RDcsan copy of R, the value of ALL_CPPFLAGS contains double quotes around one of the include paths which messes everything up.
3. CPPFLAGS="-DNDEBUG" fixes it. I think this might be the best option, because the rest of the stuff in ALL_CPPFLAGS is include dirs for R that don't pertain to libuv anyway. But I will defer to the experts.

[nealrichardson]

How about this?

diff --git a/src/Makevars b/src/Makevars
index 0dd1e62..dfe865d 100644
--- a/src/Makevars
+++ b/src/Makevars
@@ -77,7 +77,7 @@ libuv/Makefile: libuv/m4/lt~obsolete.m4
                sh autogen.sh; \
        fi; \
        chmod +x configure; \
-       CC="$(CC)" CFLAGS="$(CFLAGS) $(CPICFLAGS) $(C_VISIBILITY)" AR="$(AR)" RANLIB="$(RANLIB)" LDFLAGS="$(LDFLAGS)" ./configure $(CONFIGURE_FLAGS)
+       CC="$(CC)" CFLAGS="$(CFLAGS) $(CPICFLAGS) $(C_VISIBILITY) -DNDEBUG" AR="$(AR)" RANLIB="$(RANLIB)" LDFLAGS="$(LDFLAGS)" ./configure $(CONFIGURE_FLAGS)
 
 libuv/.libs/libuv.a: libuv/Makefile
        $(MAKE) --directory=libuv \

How would I verify in CI that the problem is resolved?

[jcheng5]

@nealrichardson That would definitely solve it--it was one of the attempts I tried yesterday

[nealrichardson]

Done in #369